Support for RDMA over Converged Ethernet is available now in Purity 5.2 software, while the ObjectEngine flash-to-flash-to-cloud data protection acquired from StorReduce will hit GA in the first half of the year.
Today. Pure Storage is making a pair of announcements, which are linked together by the way they fit within the overall grand strategy for the hybrid cloud and cloud services that the company laid out in November. They have formally announced a new DirectFlash Fabric capability for end-to-end NVMe and NVMe-oF support in the FlashArray//X Purity 5.2 software-defined engine. Out of the gate, this will provide support NVMe-oF RoCE [RDMA over Converged Ethernet], to let enterprises get flash media closer to applications for more real-time access and greater consolidation. The other announcement announces the beta launch of the technology acquired in last year’s purchase of StorReduce as a Pure Storage offering, with ObjectEngine, a data protection platform purpose-built for flash and cloud.
“This is a joint launch with two chapters of the story,” said Chadd Kenney, Vice President of Product and Solutions, Pure Storage. “At our public cloud launch in November, we laid out three core tenets for our hybrid cloud strategy using our newly introduced cloud services. They are: build your cloud, run anywhere and protect everywhere.” While our biggest announcement for 2019 is Cloud Block Store, which lets our Purity software run natively on AWS as a block storage offering, the new DirectFlash Fabric capability in Purity that supports NVMe-oF RoCE is all about building your cloud. ObjectEngine, the other part of the announcement is about the Protect Everywhere component.”
This new DirectFlash Fabric capability is all about optimizing the performance of enterprise mission-critical applications and supporting web-scale applications that traditionally have relied on lower latency Direct-Attached Storage (DAS) architectures.
“The impact of consumer trends on enterprise apps means that it is extremely rare to wait for an application to load any more,” Kenney said. “In order to deliver for this real-time society, you need to deliver high concurrency and parallelism. Flash does that incredibly well, and NVMe technology takes full advantage of that.”
Kenney said that the extension of DirectFlash to NVMe-oF is the sequel to DirectFlash Shelf, an external extension shelf for support of NVMe-over-fabrics that Pure introduced last year.
“We laid out our vision, which was to get rid of all legacy components in traditional arrays, and today Oracle now runs 3x faster, and SAP 4x faster,” he indicated. “The second part of our vision was this optimizing of the network fabric, providing full NVMe performance with all data services. It provides up to 20 per cent latency compared to Fibre Channel, which has been the best alternative until now. This is already GA, as it came out on January 16, although later this year we will get flash closer to the apps and reduce that response time even further.”
Kenney emphasized that Pure’s move directly to NVMe Fabric through RoCE rather than Fibre Channel is a strategic advantage for them.
“Many vendors plan to do NVMe-over Fabric through Fibre Channel first, but this is better because it gets the flash closer to the applications, so it provides DAS latency results with the benefits of disaggregating capacity from the compute,” he said. “It can scale compute and storage separately, and allows massive optimization in comparison to SaaS-based solutions.”
Kenney indicated that Pure will initially focus on open source applications like Cassandra and Hadoop with this.
“Next year, we will have NVMe-over TCP as well,” he added.
Kenney acknowledged that this is still an immature market, but it’s where everyone knows the industry is headed, and it’s what enterprise customers want, even though sticking with Fabric over Fibre Channel does offer much lower friction to change.
“The industry sees NVMe as the future – and not iSCSI,” he said. “People will now keep an array for 5-6, even 10 years, so they want solutions that will handle the increasing application demands. We are on the leading edge of this, showing innovation leadership, and customers will get the advantages from RDMA.”
FlashArray//X supports end-to-end NVMe on 25G and 50G Ethernet ports. Interoperability with NVMe-oF-capable NICs is available or planned from Broadcom, Cisco, Marvell and Mellanox.
The other part of the announcement is the launch of ObjectEngine into beta, with general availability planned for the first half of 2019.
“This is the industry’s first flash-to-flash-to cloud product for data protection,” said Brian Schwarz, Director of Product Management at Pure Storage. “People today try to get additional value from data for TestDev, compliance, security, and analytics workloads, but the decade-old architecture commonly in use makes it difficult. Many still use disk to disk to tape, although some have moved the cold storage to the cloud. This technology has an extremely efficient deduplication engine, and is designed to specifically address slow restore times, making restores much faster than before.”
Schwarz cited the example of a large auto part logistics company, which was hit by a ransomware attack. While all their data was protected, they were using the older technology, and the full restore of their data took seven days, with a measurable impact on their business in the meantime.
“This technology will greatly improve the speed of restores, typically by a factor or 10x, in order to meet modern SLAs,” Schwarz said.
ObjectEngine also works with backup solutions from vendors like Commvault, Veeam and Veritas – all Pure Storage partners – rather than by competing with them.
ObjectEngine will be available in two configurations. ObjectEngine//A is an appliance that delivers 25 TB/HR backup performance and 15 TB/HR restore performance, to reduce storage and bandwidth costs by up to 97 per cent. ObjectEngine//Cloud is a cloud object storage virtualization layer with a native S3 interface, that can provide 11 nines of durability in the cloud. Its near-linear scaling delivers 100+ petabyte and 100+ terabyte-per-hour performance in the cloud.
“The use case for each is determined by where the data is being generated that you need to protect,” Schwarz said. “For on-prem, you would use the appliance, and to protect cloud-native applications, it’s the cloud product.”